1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __LINUX_SEQLOCK_H
3 #define __LINUX_SEQLOCK_H
5 * Reader/writer consistent mechanism without starving writers. This type of
6 * lock for data where the reader wants a consistent set of information
7 * and is willing to retry if the information changes. There are two types
9 * 1. Sequence readers which never block a writer but they may have to retry
10 * if a writer is in progress by detecting change in sequence number.
11 * Writers do not wait for a sequence reader.
12 * 2. Locking readers which will wait if a writer or another locking reader
13 * is in progress. A locking reader in progress will also block a writer
14 * from going forward. Unlike the regular rwlock, the read lock here is
15 * exclusive so that only one locking reader can get it.
17 * This is not as cache friendly as brlock. Also, this may not work well
18 * for data that contains pointers, because any writer could
19 * invalidate a pointer that a reader was following.
21 * Expected non-blocking reader usage:
23 * seq = read_seqbegin(&foo);
25 * } while (read_seqretry(&foo, seq));
28 * On non-SMP the spin locks disappear but the writer still needs
29 * to increment the sequence variables because an interrupt routine could
30 * change the state of the data.
32 * Based on x86_64 vsyscall gettimeofday
33 * by Keith Owens and Andrea Arcangeli
36 #include <linux/spinlock.h>
37 #include <linux/preempt.h>
38 #include <linux/lockdep.h>
39 #include <linux/compiler.h>
40 #include <asm/processor.h>
43 * Version using sequence counter only.
44 * This can be used when code has its own mutex protecting the
45 * updating starting before the write_seqcountbeqin() and ending
46 * after the write_seqcount_end().
48 typedef struct seqcount {
50 #ifdef CONFIG_DEBUG_LOCK_ALLOC
51 struct lockdep_map dep_map;
55 static inline void __seqcount_init(seqcount_t *s, const char *name,
56 struct lock_class_key *key)
59 * Make sure we are not reinitializing a held lock:
61 lockdep_init_map(&s->dep_map, name, key, 0);
65 #ifdef CONFIG_DEBUG_LOCK_ALLOC
66 # define SEQCOUNT_DEP_MAP_INIT(lockname) \
67 .dep_map = { .name = #lockname } \
69 # define seqcount_init(s) \
71 static struct lock_class_key __key; \
72 __seqcount_init((s), #s, &__key); \
75 static inline void seqcount_lockdep_reader_access(const seqcount_t *s)
77 seqcount_t *l = (seqcount_t *)s;
80 local_irq_save(flags);
81 seqcount_acquire_read(&l->dep_map, 0, 0, _RET_IP_);
82 seqcount_release(&l->dep_map, 1, _RET_IP_);
83 local_irq_restore(flags);
87 # define SEQCOUNT_DEP_MAP_INIT(lockname)
88 # define seqcount_init(s) __seqcount_init(s, NULL, NULL)
89 # define seqcount_lockdep_reader_access(x)
92 #define SEQCNT_ZERO(lockname) { .sequence = 0, SEQCOUNT_DEP_MAP_INIT(lockname)}
96 * __read_seqcount_begin - begin a seq-read critical section (without barrier)
97 * @s: pointer to seqcount_t
98 * Returns: count to be passed to read_seqcount_retry
100 * __read_seqcount_begin is like read_seqcount_begin, but has no smp_rmb()
101 * barrier. Callers should ensure that smp_rmb() or equivalent ordering is
102 * provided before actually loading any of the variables that are to be
103 * protected in this critical section.
105 * Use carefully, only in critical code, and comment how the barrier is
108 static inline unsigned __read_seqcount_begin(const seqcount_t *s)
113 ret = READ_ONCE(s->sequence);
114 if (unlikely(ret & 1)) {
122 * raw_read_seqcount - Read the raw seqcount
123 * @s: pointer to seqcount_t
124 * Returns: count to be passed to read_seqcount_retry
126 * raw_read_seqcount opens a read critical section of the given
127 * seqcount without any lockdep checking and without checking or
128 * masking the LSB. Calling code is responsible for handling that.
130 static inline unsigned raw_read_seqcount(const seqcount_t *s)
132 unsigned ret = READ_ONCE(s->sequence);
138 * raw_read_seqcount_begin - start seq-read critical section w/o lockdep
139 * @s: pointer to seqcount_t
140 * Returns: count to be passed to read_seqcount_retry
142 * raw_read_seqcount_begin opens a read critical section of the given
143 * seqcount, but without any lockdep checking. Validity of the critical
144 * section is tested by checking read_seqcount_retry function.
146 static inline unsigned raw_read_seqcount_begin(const seqcount_t *s)
148 unsigned ret = __read_seqcount_begin(s);
154 * read_seqcount_begin - begin a seq-read critical section
155 * @s: pointer to seqcount_t
156 * Returns: count to be passed to read_seqcount_retry
158 * read_seqcount_begin opens a read critical section of the given seqcount.
159 * Validity of the critical section is tested by checking read_seqcount_retry
162 static inline unsigned read_seqcount_begin(const seqcount_t *s)
164 seqcount_lockdep_reader_access(s);
165 return raw_read_seqcount_begin(s);
169 * raw_seqcount_begin - begin a seq-read critical section
170 * @s: pointer to seqcount_t
171 * Returns: count to be passed to read_seqcount_retry
173 * raw_seqcount_begin opens a read critical section of the given seqcount.
174 * Validity of the critical section is tested by checking read_seqcount_retry
177 * Unlike read_seqcount_begin(), this function will not wait for the count
178 * to stabilize. If a writer is active when we begin, we will fail the
179 * read_seqcount_retry() instead of stabilizing at the beginning of the
182 static inline unsigned raw_seqcount_begin(const seqcount_t *s)
184 unsigned ret = READ_ONCE(s->sequence);
190 * __read_seqcount_retry - end a seq-read critical section (without barrier)
191 * @s: pointer to seqcount_t
192 * @start: count, from read_seqcount_begin
193 * Returns: 1 if retry is required, else 0
195 * __read_seqcount_retry is like read_seqcount_retry, but has no smp_rmb()
196 * barrier. Callers should ensure that smp_rmb() or equivalent ordering is
197 * provided before actually loading any of the variables that are to be
198 * protected in this critical section.
200 * Use carefully, only in critical code, and comment how the barrier is
203 static inline int __read_seqcount_retry(const seqcount_t *s, unsigned start)
205 return unlikely(s->sequence != start);
209 * read_seqcount_retry - end a seq-read critical section
210 * @s: pointer to seqcount_t
211 * @start: count, from read_seqcount_begin
212 * Returns: 1 if retry is required, else 0
214 * read_seqcount_retry closes a read critical section of the given seqcount.
215 * If the critical section was invalid, it must be ignored (and typically
218 static inline int read_seqcount_retry(const seqcount_t *s, unsigned start)
221 return __read_seqcount_retry(s, start);
226 static inline void raw_write_seqcount_begin(seqcount_t *s)
232 static inline void raw_write_seqcount_end(seqcount_t *s)
239 * raw_write_seqcount_barrier - do a seq write barrier
240 * @s: pointer to seqcount_t
242 * This can be used to provide an ordering guarantee instead of the
243 * usual consistency guarantee. It is one wmb cheaper, because we can
244 * collapse the two back-to-back wmb()s.
246 * Note that, writes surrounding the barrier should be declared atomic (e.g.
247 * via WRITE_ONCE): a) to ensure the writes become visible to other threads
248 * atomically, avoiding compiler optimizations; b) to document which writes are
249 * meant to propagate to the reader critical section. This is necessary because
250 * neither writes before and after the barrier are enclosed in a seq-writer
251 * critical section that would ensure readers are aware of ongoing writes.
254 * bool X = true, Y = false;
261 * int s = read_seqcount_begin(&seq);
265 * } while (read_seqcount_retry(&seq, s));
272 * WRITE_ONCE(Y, true);
274 * raw_write_seqcount_barrier(seq);
276 * WRITE_ONCE(X, false);
279 static inline void raw_write_seqcount_barrier(seqcount_t *s)
286 static inline int raw_read_seqcount_latch(seqcount_t *s)
288 /* Pairs with the first smp_wmb() in raw_write_seqcount_latch() */
289 int seq = READ_ONCE(s->sequence); /* ^^^ */
294 * raw_write_seqcount_latch - redirect readers to even/odd copy
295 * @s: pointer to seqcount_t
297 * The latch technique is a multiversion concurrency control method that allows
298 * queries during non-atomic modifications. If you can guarantee queries never
299 * interrupt the modification -- e.g. the concurrency is strictly between CPUs
300 * -- you most likely do not need this.
302 * Where the traditional RCU/lockless data structures rely on atomic
303 * modifications to ensure queries observe either the old or the new state the
304 * latch allows the same for non-atomic updates. The trade-off is doubling the
305 * cost of storage; we have to maintain two copies of the entire data
308 * Very simply put: we first modify one copy and then the other. This ensures
309 * there is always one copy in a stable state, ready to give us an answer.
311 * The basic form is a data structure like:
313 * struct latch_struct {
315 * struct data_struct data[2];
318 * Where a modification, which is assumed to be externally serialized, does the
321 * void latch_modify(struct latch_struct *latch, ...)
323 * smp_wmb(); <- Ensure that the last data[1] update is visible
325 * smp_wmb(); <- Ensure that the seqcount update is visible
327 * modify(latch->data[0], ...);
329 * smp_wmb(); <- Ensure that the data[0] update is visible
331 * smp_wmb(); <- Ensure that the seqcount update is visible
333 * modify(latch->data[1], ...);
336 * The query will have a form like:
338 * struct entry *latch_query(struct latch_struct *latch, ...)
340 * struct entry *entry;
344 * seq = raw_read_seqcount_latch(&latch->seq);
347 * entry = data_query(latch->data[idx], ...);
350 * } while (seq != latch->seq);
355 * So during the modification, queries are first redirected to data[1]. Then we
356 * modify data[0]. When that is complete, we redirect queries back to data[0]
357 * and we can modify data[1].
359 * NOTE: The non-requirement for atomic modifications does _NOT_ include
360 * the publishing of new entries in the case where data is a dynamic
363 * An iteration might start in data[0] and get suspended long enough
364 * to miss an entire modification sequence, once it resumes it might
365 * observe the new entry.
367 * NOTE: When data is a dynamic data structure; one should use regular RCU
368 * patterns to manage the lifetimes of the objects within.
370 static inline void raw_write_seqcount_latch(seqcount_t *s)
372 smp_wmb(); /* prior stores before incrementing "sequence" */
374 smp_wmb(); /* increment "sequence" before following stores */
378 * Sequence counter only version assumes that callers are using their
381 static inline void write_seqcount_begin_nested(seqcount_t *s, int subclass)
383 raw_write_seqcount_begin(s);
384 seqcount_acquire(&s->dep_map, subclass, 0, _RET_IP_);
387 static inline void write_seqcount_begin(seqcount_t *s)
389 write_seqcount_begin_nested(s, 0);
392 static inline void write_seqcount_end(seqcount_t *s)
394 seqcount_release(&s->dep_map, 1, _RET_IP_);
395 raw_write_seqcount_end(s);
399 * write_seqcount_invalidate - invalidate in-progress read-side seq operations
400 * @s: pointer to seqcount_t
402 * After write_seqcount_invalidate, no read-side seq operations will complete
403 * successfully and see data older than this.
405 static inline void write_seqcount_invalidate(seqcount_t *s)
412 struct seqcount seqcount;
417 * These macros triggered gcc-3.x compile-time problems. We think these are
418 * OK now. Be cautious.
420 #define __SEQLOCK_UNLOCKED(lockname) \
422 .seqcount = SEQCNT_ZERO(lockname), \
423 .lock = __SPIN_LOCK_UNLOCKED(lockname) \
426 #define seqlock_init(x) \
428 seqcount_init(&(x)->seqcount); \
429 spin_lock_init(&(x)->lock); \
432 #define DEFINE_SEQLOCK(x) \
433 seqlock_t x = __SEQLOCK_UNLOCKED(x)
436 * Read side functions for starting and finalizing a read side section.
438 static inline unsigned read_seqbegin(const seqlock_t *sl)
440 return read_seqcount_begin(&sl->seqcount);
443 static inline unsigned read_seqretry(const seqlock_t *sl, unsigned start)
445 return read_seqcount_retry(&sl->seqcount, start);
449 * Lock out other writers and update the count.
450 * Acts like a normal spin_lock/unlock.
451 * Don't need preempt_disable() because that is in the spin_lock already.
453 static inline void write_seqlock(seqlock_t *sl)
455 spin_lock(&sl->lock);
456 write_seqcount_begin(&sl->seqcount);
459 static inline void write_sequnlock(seqlock_t *sl)
461 write_seqcount_end(&sl->seqcount);
462 spin_unlock(&sl->lock);
465 static inline void write_seqlock_bh(seqlock_t *sl)
467 spin_lock_bh(&sl->lock);
468 write_seqcount_begin(&sl->seqcount);
471 static inline void write_sequnlock_bh(seqlock_t *sl)
473 write_seqcount_end(&sl->seqcount);
474 spin_unlock_bh(&sl->lock);
477 static inline void write_seqlock_irq(seqlock_t *sl)
479 spin_lock_irq(&sl->lock);
480 write_seqcount_begin(&sl->seqcount);
483 static inline void write_sequnlock_irq(seqlock_t *sl)
485 write_seqcount_end(&sl->seqcount);
486 spin_unlock_irq(&sl->lock);
489 static inline unsigned long __write_seqlock_irqsave(seqlock_t *sl)
493 spin_lock_irqsave(&sl->lock, flags);
494 write_seqcount_begin(&sl->seqcount);
498 #define write_seqlock_irqsave(lock, flags) \
499 do { flags = __write_seqlock_irqsave(lock); } while (0)
502 write_sequnlock_irqrestore(seqlock_t *sl, unsigned long flags)
504 write_seqcount_end(&sl->seqcount);
505 spin_unlock_irqrestore(&sl->lock, flags);
509 * A locking reader exclusively locks out other writers and locking readers,
510 * but doesn't update the sequence number. Acts like a normal spin_lock/unlock.
511 * Don't need preempt_disable() because that is in the spin_lock already.
513 static inline void read_seqlock_excl(seqlock_t *sl)
515 spin_lock(&sl->lock);
518 static inline void read_sequnlock_excl(seqlock_t *sl)
520 spin_unlock(&sl->lock);
524 * read_seqbegin_or_lock - begin a sequence number check or locking block
525 * @lock: sequence lock
526 * @seq : sequence number to be checked
528 * First try it once optimistically without taking the lock. If that fails,
529 * take the lock. The sequence number is also used as a marker for deciding
530 * whether to be a reader (even) or writer (odd).
531 * N.B. seq must be initialized to an even number to begin with.
533 static inline void read_seqbegin_or_lock(seqlock_t *lock, int *seq)
535 if (!(*seq & 1)) /* Even */
536 *seq = read_seqbegin(lock);
538 read_seqlock_excl(lock);
541 static inline int need_seqretry(seqlock_t *lock, int seq)
543 return !(seq & 1) && read_seqretry(lock, seq);
546 static inline void done_seqretry(seqlock_t *lock, int seq)
549 read_sequnlock_excl(lock);
552 static inline void read_seqlock_excl_bh(seqlock_t *sl)
554 spin_lock_bh(&sl->lock);
557 static inline void read_sequnlock_excl_bh(seqlock_t *sl)
559 spin_unlock_bh(&sl->lock);
562 static inline void read_seqlock_excl_irq(seqlock_t *sl)
564 spin_lock_irq(&sl->lock);
567 static inline void read_sequnlock_excl_irq(seqlock_t *sl)
569 spin_unlock_irq(&sl->lock);
572 static inline unsigned long __read_seqlock_excl_irqsave(seqlock_t *sl)
576 spin_lock_irqsave(&sl->lock, flags);
580 #define read_seqlock_excl_irqsave(lock, flags) \
581 do { flags = __read_seqlock_excl_irqsave(lock); } while (0)
584 read_sequnlock_excl_irqrestore(seqlock_t *sl, unsigned long flags)
586 spin_unlock_irqrestore(&sl->lock, flags);
589 static inline unsigned long
590 read_seqbegin_or_lock_irqsave(seqlock_t *lock, int *seq)
592 unsigned long flags = 0;
594 if (!(*seq & 1)) /* Even */
595 *seq = read_seqbegin(lock);
597 read_seqlock_excl_irqsave(lock, flags);
603 done_seqretry_irqrestore(seqlock_t *lock, int seq, unsigned long flags)
606 read_sequnlock_excl_irqrestore(lock, flags);
608 #endif /* __LINUX_SEQLOCK_H */